Dimyadi, J., Fernando, S., Davies, K., & Amor, R.
2020
Read Publication
One key ingredient in the automated compliance audit process is the availability of a computable form of normative requirements (e.g. codes and standards), which are usually written in natural language intended for human interpretation and not readily processable by machines.
The predominantly ‘Blackbox’ approach of hardcoding these computable normative rules into a compliance audit system has been reported to be problematic and costly to maintain in response to frequent regulatory changes. The current research sets out to investigate to what extent normative texts can be represented as computable rules for automated compliance audit as well as to ease maintenance in response to changes in the source documents. A set of priority compliance documents supporting the New Zealand Building Code has been selected as the subject for a case study. This paper describes the digitisation and quality assurance process, the knowledge extraction experience, and challenges identified during the study. Furthermore, the paper explores how the legal knowledge captured by the digitised rules can be used effectively in an automated compliance audit environment. The findings from the study suggest that a semi-automated digitisation process is feasible and up to 80% of prescriptive text can be translated and encoded into the open standard LegalRuleML. However, only approximately 50% of these can be used directly in an automated compliance audit environment without any human intervention. The lessons learnt from the study can be used towards improving the digitisation process. Ultimately, this could in turn help to improve the natural language source text in subsequent revisions of the codes.
